Digital watermark technique refers to use the signal processing technique to embed a hidden mark in digital multimedia data, and this kind of marks is usually invisible and only can be extracted through a dedicated detector or reader. The information embedded in the digital media can be called as the digital watermark only if the information embedded in the digital media has following two basic properties to be able to called digital watermark:                (a) invisibility: embedding the digital watermark in digital media will not degrade the quality evidently, and it is unnoticeable;        (b) robustness: robustness means the digital watermark can still maintain the integrity or can be accurately identified after experiencing a variety of unintentional or intentional signal processing. Possible signal processing comprises channel noise, filtering, digital/analog conversion, analog/digital conversion, re-sampling, shearing, displacement, scale changing, and lossy compression coding and so on.        
In current various techniques for embedding and extracting the watermark, the transform domain watermark becomes a more important watermark technique due to the advantage that it can embed a large amount of data without reducing the invisibility of the watermark. Such technique is generally based on the common image transforms and based on partial or full transforms, and these transforms include the Discrete Cosine Transform (DCT), Discrete Wavelet transform (DWT), Discrete Fourier Transform (DFT), Singular Value Decomposition (SVD), and Discrete Hadamard Transform (DHT) and so on, wherein the DCT based on the block is one of the most commonly used transform, and the still image compression standard JPEG also based on the block DCT.
One of the earliest digital watermark techniques based on the block DCT is that a key randomly selects a plurality of blocks in an image, and a triple is slightly changed in the intermediate frequency coefficients of the frequency domain coefficients to hide binary sequence information. Selecting to code the intermediate frequency component is because the high frequency coding is easily damaged by a variety of signal processing methods, and the low frequency coding changes are easily noticeable since human vision is very sensitive to low frequency components. The digital watermark technique is robust to the lossy compression and the low pass filtering.
Another DCT digital watermark technique is first dividing the image into 8×8 non-overlapped blocks, obtaining frequency blocks composed by DCT coefficients after dividing into blocks of DCT, then randomly selecting a plurality of frequency blocks, and embedding the watermark signal into a plurality of DCT coefficients selected under the control of the key. The method slightly changes the selected DCT coefficients to meets the specific relationship so as to represent information of one bit. When extracting the watermark information, the same DCT coefficients are selected and the bit information is extracted according to the relationship between the coefficients.
Besides the abovementioned representative transform domain methods, there are further a plurality of digital watermark techniques based on the transform domain, a considerable portion of these techniques are the improvement and development of the abovementioned methods, wherein a representative method is the one proposed by I. Podichuk and Zeng Wen-jun. The method is based on the DCT or wavelet transform of still images and studies where the digital watermark returned by the vision model module should be loaded and the affordable Just Noticeable Difference (JND) value (the intensity upper limit of loading the watermark) at each place, and this watermark method is adaptive.
To date, most digital watermarks based on the transform domain are private watermarks, in which the original digital media is needed so that the digital watermarks can be detected, and the private watermarks can only be used as copyright protection, which largely limits the application range of digital watermark techniques. Besides, the existing digital watermark techniques only use a single transform to embed the watermarks, which makes the anti-attack ability of the watermarks greatly reduced.